Population pharmacokinetics of vancomycin and AUC-guided dosing in Chinese neonates and young infants

  • Yewei Chen
  • Dan Wu
  • Min Dong
  • Yiqing Zhu
  • Jinmiao Lu
  • Xiaoxia Li
  • Chao Chen
  • Zhiping Li
Pharmacokinetics and Disposition
  • 100 Downloads

Abstract

Objectives

To develop a population pharmacokinetic (PK) model for vancomycin in Chinese neonates and infants less than 2 months of age (young infants) with a wide gestational age range, in order to determine the appropriate dosing regimen for this population.

Methods

We performed a retrospective chart review of patients from the neonatal intensive care unit (NICU) at Children’s Hospital of Fudan University to identify neonates and young infants treated with vancomycin from May 2014 to May 2017. Vancomycin concentrations and covariates were utilized to develop a one-compartment model with first-order elimination. The predictive performance of the final model was assessed by both internal and external evaluation, and the relationship between trough concentration and AUC0–24 was investigated. Monte Carlo simulations were performed to design an initial dosing schedule targeting an AUC0–24 ≥ 400.

Results

The analysis included a total of 330 concentration–time data points from 213 neonates and young infants with gestational age (GA) and body weight of 25–42 weeks and 0.88–5.1 kg, respectively. Body weight, postmenstrual age (PMA) and serum creatinine level were found to be important factors explaining the between-subject variability in vancomycin PK parameters for this population. Both internal and external evaluation supported the prediction of the final vancomycin PK model. The typical population parameter estimates of clearance and distribution volume for an infant weighing 2.73 kg with a PMA of 39.8 weeks and serum creatinine of 0.28 mg/dL were 0.103 L/h/kg and 0.58 L/kg, respectively. Although vancomycin serum trough concentrations were predictive of the AUC, considerable variability was observed in the achievement of an AUC0–24/MIC of ≥400. For MIC values of ≤0.5 mg/L, AUC0–24/MIC ≥400 was achieved for 95% of the newborn infants with vancomycin troughs of 5–10 mg/L. When the MIC increased to 1 mg/L, only 15% of the patients with troughs of 5–10 mg/L achieved AUC0–24/MIC ≥400. For MIC values of 2 mg/L, no infants achieved the target. Simulations predicted that a dose of at least 14 and 15 mg/kg every 12 h was required to attain the target AUC0–24 ≥ 400 in 90% of infants with a PMA of 30–32 and 32–34 weeks, respectively. This target was also achieved in 93% of simulated infants in the oldest PMA groups (36–38 and 38–40 weeks, respectively) when the dosing interval was extended to 8 h. For infants with a PMA ≥44 weeks, a dose increase to 18 mg/kg every 8 h was needed. The trough concentrations of 5–15 mg/L were highly predictive of an AUC0–24 of ≥400 when treating invasive MRSA infections with an MIC of ≤1 mg/L.

Conclusions

The PK parameters for vancomycin in Chinese infants younger than 2 months of age were estimated using the model developed herein. This model has been used to predict individualized dosing regimens in this vulnerable population in our hospital. A large external evaluation of our model will be conducted in future studies.

Keywords

Vancomycin Pharmacokinetics NONMEM AUC Infants 

Introduction

Vancomycin is the first-choice antibiotic for newborn infants with suspected or confirmed β-lactam-resistant gram-positive bacterial infections, such as methicillin-resistant Staphylococcus aureus (MRSA) [1]. Although it has been prescribed for more than 50 years and has been widely studied, many questions remain about the optimal and safe use of vancomycin in newborn infants [2]. Underestimating true vancomycin exposure may lead to an increased risk of renal toxicity, while overestimation may be associated with treatment failure, which underscores the importance of optimizing vancomycin dosing in neonates and young infants to rapidly achieve adequate drug exposure. However, this has been challenging, as the pharmacokinetics of vancomycin are highly variable in this population due to differences in maturation and development [3, 4]. Although clinicians still rely on trough concentration monitoring, which is dependent on the dosing interval, data from experimental and clinical studies have shown that the best metric for optimization of vancomycin dosing in this population is the ratio of the area under the 24-h concentration–time curve to the minimum inhibitory concentration (AUC0–24/MIC) [5, 6, 7, 8].

Targeting an AUC0–24/MIC ≥400 is recommended by the Infectious Disease Society of America when treating invasive MRSA infections, and a trough concentration of 15–20 mg/L is generally required to achieve this target when the MIC is ≤1 mg/L (as is common for many strains of MRSA) in adults [9]. This strong correlation between trough and AUC0–24/MIC to some extent reduces the reliance on calculating AUC0–24 in adults. However, the relationship between trough and AUC0–24/MIC in adults may not extrapolate to pediatrics [10, 11, 12]. Lanke et al. reported that troughs ranging from 10 to 12.5 mg/L were highly predictive of achieving an AUC0–24/MIC ≥400 in adolescents when the MIC was ≤1 mg/L [13]. Frymoyer et al. suggest that a vancomycin trough concentration of 15–20 mg/L is unnecessary, and that lower trough concentrations of 7–10 mg/L should be sufficient for treatment of invasive MRSA infection in >90% of neonates when the MIC of vancomycin is 1 mg/L [12, 14]. Le et al. found that an AUC0–24/MIC of ≥400 correlated with a similar trough concentration of 8–9 mg/L in 75% of pediatric patients [15]. The lower target trough concentration raises important topics for discussion. Hahn et al. performed a validation study of a pediatric population pharmacokinetic (PK) model of vancomycin at their institution, and they do not recommend a lower trough concentration, but advise individual AUC estimation using Bayesian approaches [16]. Although researchers have recently proposed novel dosing guidelines for vancomycin in neonates based on AUC targets [17], it is still unknown whether the previous dosing advice is applicable to Chinese populations.

Population PK models are a powerful tool that can aid clinicians and facilitate individualized drug therapy. By incorporating patient-specific characteristics, dosing information, drug concentrations, and consideration of intra- and inter- patient variability, population PK models can provide a more personalized approach to therapeutic decision-making [18, 19]. The purpose of this investigation was to establish a population PK model of vancomycin in Chinese neonates and infants less than 2 months of age (“young infants”). In addition, an external evaluation was performed to test the predictive performance in an independent dataset. Additionally, the obtained population PK parameters were used to predict dosing regimens in neonates and young infants that generated a high likelihood of achieving an AUC0–24/MIC ≥400.

Methods

Patients and data collection

To construct the model, data were collected from neonates and young infants in the neonatal intensive care unit (NICU) at Children’s Hospital of Fudan University from May 2014 to May 2017. Additional patients with similar characteristics in our NICU from June 2017 to December 2017 were enrolled for model evaluation. The study protocol was approved by the ethics committee of our hospital. Inclusion criteria were as follows: age <61 days, sufficient intravascular access (either peripheral or central) to receive the study drug, and suspected or confirmed gram-positive infection that necessitated treatment with vancomycin as part of the standard of care following the decision of the attending physician. Participants were excluded if a complete vancomycin dosing history was not available, a diagnosis of congenital kidney disease or major congenital heart disease was made, a concentration was below the lower limit of quantification (LLOQ), or any condition (receiving hemodialysis or extracorporeal membrane oxygenation during vancomycin administration) was present which, in the opinion of the investigator, made the subject unsuitable for enrollment.

For those meeting the enrollment criteria, the following demographic and biochemical factors were collected: gestational age (GA), postnatal age (PNA), postmenstrual age (PMA), birth body weight (BBW), body weight (BW), height (HT), body surface area [BSA, \( BSA\left({m}^2\right)=\sqrt{\frac{HT(cm)\ast BW(kg)}{3600}} \)], serum creatinine (Scr), creatinine clearance rate [CLcr, CLcr(mL/ min /1.73m2) = k ∗ HT(cm)/Scr(mg/dL), where k = 0.45 for term infants throughout the first year of life] [20], and concomitant drug therapy.

For the representative MIC of MRSA isolates, pediatric cultures at our hospital from 2014 to 2017 were reviewed.

Dosing regimen and sampling

The initial vancomycin dosage regimen for neonates and young infants was 10 or 15 mg/kg (for bacteremia or meningitis, respectively) every 12 or 8 h, based on combinations of PMA and PNA as listed in the Neofax® manual [21], with further dosing guided by therapeutic drug motoring (TDM). PMA is the primary determinant of dosing interval, with PNA the secondary qualifier. Vancomycin was administered intravenously over 60 min. Blood samples were collected from all patients for TDM as part of routine medical care. The trough concentration was obtained 30 min prior to the fourth dose, and the peak concentration was obtained 1 h after the initiation of the 1-h infusion.

Assay of serum vancomycin

The serum vancomycin trough and peak concentrations were determined using an enzyme-multiplied immunoassay method with the Viva-E System (ver. 2.014; Siemens Healthcare Diagnostics, Eschborn, Germany). The linear range for the assay was 2.0–50 mg/L, and the LLOQ was 2 mg/L.

Pharmacokinetic analysis

Data analysis was performed with the NONMEM software program (ver. VII; Icon Development Solutions, Ellicott City, MD, USA), in the R programming environment (ver. 14.2, http://www.r-project.org/). The first-order conditional estimation (FOCE) method with interaction option was used to estimate PK parameters and their variability. A one-compartment model with first-order elimination was used as the PK base model. The one-compartment PK parameters were clearance (CL) and apparent volume of distribution (V). Interindividual variability was evaluated on CL and V using an exponential model. To model the residual variability, both additive and proportional error models were evaluated.

The demographic information was used to perform an initial selection of covariates. The selection was carried out by plotting the parameter estimates against demographic factors, and retaining those with statistical significance as initial covariates. It is well known that size (weight) and maturation (age) have a significant impact on the CL of vancomycin in newborn infants [22, 23]. Therefore, weight and age were included in the model first. Because weight was proven to be superior to age as a covariate for CL, models based on BW scaling of the intrinsic clearance were tested using Eq. 1.
$$ CL={CL}_{std}\cdot {\left(\frac{BW}{70}\right)}^{k1} $$
(1)
where CLstd represents the clearance in an adult with a body weight of 70 kg, and k1 is the exponent. Allometric scaling is the most widely used to describe size differences and has a strong theoretical basis with an exponent of 0.75 for clearance. The exponent k1 was fixed to 0.75. Gestational age (GA), postnatal age (PNA), and postmenstrual age (PMA) with linear, exponential and sigmoid Emax maturation function were tested to explore the effect of maturational changes on vancomycin CL [23]. The effects of serum creatinine and creatinine clearance on vancomycin CL were modeled, respectively, assuming an exponential relationship. The covariate model of CL with creatinine clearance was finalized with the addition of only PMA (not weight). It is assumed that weight is already included in creatinine clearance.

The selection of covariates was determined using a forward selection process and a backward elimination process. Nested models were statistically compared using a likelihood ratio test on the differences in the objective function value (OFV). A reduction in OFV of 3.84 (p < 0.05) for forward inclusion and an increase in OFV of 10.83 (p < 0.001) for backward elimination were the criteria for retaining a covariate in the model. The Akaike information criterion (AIC), calculated using Pirana software (ver. 2.7.1; Pirana Software & Consulting BV, http://www.pirana-software.com/), was used to select competing, non-nested models. Models with lower AICs were considered superior.

Model evaluation

The goodness of fit was evaluated using several diagnostic scatter plots, as follows: (1) observed versus population-predicted concentrations (DV vs. PRED); (2) observed versus individual predicted concentrations (DV vs. IPRED); (3) conditional weighted residuals versus time (CWRES vs. TIME); (4) conditional weighted residuals versus population-predicted concentrations (CWRES vs. PRED).

The accuracy and stability of the final model was assessed by means of an internal evaluation method involving a non-parametric bootstrap [24]. Re-sampling of the data set was carried out 1000 times using NONMEM in the final model. The values of estimated parameters such as the median and SE from the bootstrap procedure were compared with those estimated from the original data set. The model was proven to be stable if the values of the parameters were not significantly different.

The normalized prediction distribution error (NPDE) was calculated to evaluate the predictive performance of the model [25]. One thousand data sets were simulated based on the final model. NPDE results were summarized graphically using the NPDE R package. The NPDE distribution was expected to follow a normal distribution (mean, 0; variance, 1).

To perform an external evaluation for the final vancomycin model, we enrolled an additional 57 neonates and infants in the study. The population-predicted concentrations were compared to the observed concentrations using the MAXEVAL = 0 option in NONMEM. The predictive performance of the final model was evaluated by means of precision and bias. The mean prediction error (MPE) and mean absolute prediction error (MAPE) were used as measures of precision and bias [26]. These are calculated by the following equations:
$$ {\displaystyle \begin{array}{c} MPE\%=\frac{1}{N}\sum \frac{PRED_i-{OBS}_i}{OBS_i}\times 100\%\\ {} MAPE\%=\frac{1}{N}\sum \left|\frac{PRED_i-{OBS}_i}{OBS_i}\right|\times 100\%\end{array}} $$
where OBSi represents the observed concentration of the ith subject, and PREDi represents the population-predicted concentration of the ith subject. In addition, the final model was used to calculate the number of patients with MPE within ±20% and ±30%. The final model with low MPE and MAE values and a high number of prediction errors within ±20% and ±30% was considered acceptable.

Assessment of trough concentration and AUC0–24 relationship

One thousand data sets were simulated based on the final model. Serum concentrations were obtained every 1 h after steady-state achievement for each patient (after four doses had been administered). The AUC0–24 was estimated from 0 to 24 h using the trapezoidal rule in GraphPad Prism (ver. 4.0; GraphPad Software Inc., San Diego, CA, USA). The relationships between observed trough concentration and the proportion of subjects who achieved an AUC0–24/MIC ≥400 (for 0.5 mg/L, 1 mg/L or 2 mg/L MIC) were then examined.

Dosing optimization

The final PK model was utilized to identify a dosing regimen that could reliably achieve the targeted ratio of an AUC0–24/MIC ≥400. Monte Carlo simulations were used to simulate vancomycin exposure for 1000 subjects randomly selected from a database. The database contains information for all neonates and young infants with suspected gram-positive infection (n = 21,078) discharged from our NICU from 2014 to 2017. The analysis was approved by the ethics committee of our hospital without the need for written informed consent, since the data were collected without patient identifiers. Demographic ranges for BW, PMA, and SCR were included when generating the random sample to match demographic distribution of study subjects. Random numbers were generated from the uniform distribution using Excel (ver. 2013; Microsoft Corporation, Redmond, WA, USA). The proportion of simulated subject profiles that met an AUC0–24 ≥ 400 was calculated for each dosing recommendation. In view of the efficacy and toxicity in relation to concentrations, trough concentrations were predicted when ≥90% of predicted AUC0–24 was ≥ 400.

Results

Data analysis

A total of 213 patients were enrolled in the study. The patient characteristics are presented in Table 1. The final PK database consisted of 330 vancomycin concentrations. Nine patients were excluded due to LLOQ.
Table 1

Demographic data of patients

Characteristica

Model-building data setb

External data setb

No. of patients/samples

213/330

57/64

No. of trough/peak concentrations

213/117

57/7

GA (weeks)

36.9 [25–42]

38.7 [26.9–42]

PNA (days)

26 [6–59]

14 [0–43]

PMA (weeks)

39.8 [28–47.9]

40.7 [29.2–46.4]

BBW (kg)

2.53 [0.7–4.7]

3 [1.1–5]

BW (kg)

2.73 [0.88–5.1]

2.9 [0.8–6]

HT (cm)

49 [28–57]

49 [27–57]

BSA (m2)

0.19 [0.09–0.28]

0.20 [0.09–0.27]

Scr (mg/dL)

0.28 [0.11–0.72]

0.34 [0.15–0.75]

CLcr (mL/min/1.73 m2)

68 [17–219]

57 [22–153]

aGA, gestational age; PNA, postnatal age; PMA, postmenstrual age; BBW, birth body weight; BW, body weight; HT, height; BSA, body surface area; Scr, serum creatinine; CLcr, creatinine clearance rate, CLcr(mL/ min /1.73m2) = k ∗ HT(cm)/Scr(mg/dL)

bThe data are presented as median [range]

Population PK modeling

Preliminary analysis for the base model showed that the OFV of the one-compartment model was 1545.04. Residual variability was best described by a combined proportional and additive error model.

To account for size and maturation, six equations for clearance were evaluated (Table 2). Among the six models examined, the 0.75 allometric and sigmoidal model had the lowest AIC (1288.637) and was employed for further covariate analysis. Postmenstrual age was superior to postnatal age. The inclusion of weight and PMA for the prediction of CL in the population PK model resulted in an ΔOFV of −270.403 (p < 0.01). The addition of weight scaled by allometry to V also improved the model (ΔOFV, −34.98; p < 0.01). Although weight and PMA were included in the model, the serum creatinine level was a significant predictor of vancomycin (ΔOFV, −13.716; p < 0.01). The inclusion of weight, PMA and serum creatinine to CL and weight to V produced the most significant decrease in the OFV (ΔOFV, −319.099) and between-subject variability (BSVS) for CL (ωCL decreased from 0.61 to 0.27). All covariates passed the backward elimination criteria. The final vancomycin population PK model, including parameter estimates and their relative standard errors (RSE), are given in Table 3.
Table 2

Effect of covariate analysis—impact of each covariate when added sequentially to the model

Covariates

Model description

OFV

AIC

ΔAIC

ωCL

η-Shrinkage (%)

Base model

One-compartment

1545.04

1563.04

0.000

0.61 (12.3)

7.2

Step 1: Size

 3/4 Allometric model

CLstd ⋅ (BW/70)0.75

1358.671

1376.671

−186.369

0.34 (14.8)

12.6

 BW exponential model

CLstd ⋅ (BW/70)k1

1311.317

1339.317

−223.723

0.54 (19.6)

17.7

Maturation

Step 2: 3/4 Allometric model based

 Sigmoidal model

CLstd ⋅ (BW/70)0.75 ⋅ FMAT

1274.637

1288.637

−274.403

0.27 (19.7)

17.8

 PMA exponential model

CLstd ⋅ (BW/70)0.75 ⋅ (PMA/39.8) EXP

1291.792

1303.792

−259.248

0.28 (19.0)

17.5

 PNA exponential model

CLstd ⋅ (BW/70)0.75 ⋅ (PNA/26) EXP

1354.676

1366.676

−196.364

0.28 (15.7)

12.5

BW exponential model based

 Sigmoidal model

CLstd ⋅ (BW/70)k1 ⋅ FMAT

1285.336

1301.336

−261.704

0.28 (19.3)

17.7

 PMA exponential model

CLstd ⋅ (BW/70)k1 ⋅ (PMA/39.8) EXP

1291.790

1333.79

−229.25

0.28 (18.7)

17.5

 PNA exponential model

CLstd ⋅ (BW/70)k1 ⋅ (PNA/26) EXP

1300.833

1314.833

−248.207

0.29 (19.5)

17.8

Step 3: V

 3/4 Allometric model

CLstd ⋅ (BW/70)0.75 ⋅ FMAT

V ⋅ (BW/70)

1239.657

1253.657

−309.383

0.28 (16.2)

12.5

Step 4: Renal function

 SCR exponential model

CLstd ⋅ (BW/70)0.75 ⋅ FMAT ⋅ (SCR/0.28) EXP V ⋅ (BW/70)

1225.941

1241.941

−321.099

0.27 (16.6)

12.8

OFV, objective function value; AIC, Akaike information criterion; ΔAIC, change in Akaike information criterion; ωCL, inter-subject variability of clearance; CLstd, clearance in an adult with a body weight of 70 kg; BW, body weight; PMA, postmenstrual age; PNA, postnatal age; SCR, serum creatinine; k1, exponent coefficient of BW; EXP, exponential function

FMAT = PMA HillCL /(PMA HillCL  + TM50 HillCL ), where TM50 is the PMA at which clearance is 50% of that of the mature value, and HillCL is the Hill coefficient for clearance

Table 3

Parameter estimates of the final vancomycin model and bootstrap validation

Parameter

Final model

Bootstrap n = 1000

Population estimate

RSE (%)

Median

95% CI

Clearance (L/h): CL = θ1*(BW/70)**0.75*(PMA**θ4/(PMA**θ4 + θ3**θ4))*(SCR/0.28)** θ5

 θ1

4.87

19.5

4.86

3.82–11.7

 θ3

34.5

9.4

34.4

31.1–59.6

 θ4

4.61

27.1

4.66

2.45–7.41

 θ5

−0.221

27.8

−0.223

−0.345 to −0.099

Volume of distribution (L): V = θ2*(BW/70)

 θ2

40.7

4.6

40.8

37.2–44.8

Inter-individual variability

 CL (%CV)

26.8

16.6

26.4

22.1–31

Residual error model

 Proportional (%CV)

23.9

12.4

23.6

20.8–26.4

 Additive (mg/L)

0.688

44.9

0.659

0.397–1

RSE, relative standard error; CI, confidence interval

Model evaluation

Diagnostic plots for the final vancomycin model showed a good model fit (Fig. 1). The results of 1000 bootstrap replicates for vancomycin are summarized in Table 3. The number of runs successfully converged was 951. The median parameter estimates from the bootstrap procedure were very close to the values in the final population model. In addition, the parameters from the bootstrap procedure followed a normal distribution and contained all of the parameter estimates from the final population model. The results indicate that the estimates for the population PK parameters in the final model were precise and that the model was stable. An internal model evaluation also demonstrated that the final model performed well in describing the observed data (Fig. 2). The mean NPDE was 0.07 (theoretical mean is zero), and there were no trends in NPDE across time or predicted vancomycin concentrations.
Fig. 1

Diagnostic plots of the vancomycin final model. a The observed versus population-predicted concentration. b The observed versus individual-predicted concentration. c Conditional weighted residual versus time. d Conditional weighted residual versus the predicted concentration

Fig. 2

Normalized prediction distribution errors (NPDE) of the vancomycin final model. a Q–Q plot of the NPDE. b Histogram of the NPDE. c NPDE versus time after first dose. d NPDE versus population-predicted concentration (PRED)

The external evaluation suggested that the final model accurately characterized the PK profile of vancomycin in the population. Demographic and clinical characteristics of the neonates and young infants in the external evaluation are shown in Table 1. The plot of observed vancomycin concentrations versus population-predicted concentrations is shown in Fig. 3. The MPE and APE were 12.3% and 31%, respectively. The percentage of population prediction error within ±20% and ±30% was 42.2% and 62.5%, respectively. All validation parameters indicated good predictive performance of the model in new patients.
Fig. 3

Observed vancomycin concentrations versus population-predicted concentrations for the external evaluation data set

Assessment of trough concentration and AUC0–24 relationship

Across the 213 neonates and young infants, the median AUC0–24 was 299 (range, 142 to 659) mg·/h/L. The simulation results indicate that the current dosing regimens based on Neofax® produced effective therapeutic exposure for patients in our hospital (where the MIC values for MRSA are 0.5 mg/L or less). A comparison was performed between trough concentrations and AUC0–24 (Fig. 4). For MIC values of 0.5 mg/L, AUC0–24/MIC ≥400 was achieved for 88% (56/64), 95% (111/117) and 100% (32/32) of the newborn infants with vancomycin troughs <5 mg/L, 5–10 mg/L and > 10 mg/L, respectively. When the MIC increased to 1 mg/L, target attainment dropped to 6% (4/64), 15% (17/117), 14% (3/21) and 55% (6/11) of patients with trough concentrations of <5 mg/L, 5–10 mg/L, 10–15 mg/L and > 15 mg/L, respectively. For MIC values of 2 mg/L, no infants achieved AUC0–24/MIC ≥400. It is worth noting that the lower vancomycin MIC in our hospital helped to account for the increased numbers of patients achieving an AUC0–24/MIC ≥400.
Fig. 4

Scatter plot of initial vancomycin serum trough concentrations and AUC0–24 (blue, black, pink and green represent troughs of <5 mg/L, 5-10 mg/L, 10-15 mg/L and > 15 mg/L, respectively)

Dosing optimization

Monte Carlo simulations performed with virtual patients randomly selected from a database showed that the AUC-guided dosing regimen achieved AUC0–24 of ≥400 in ≥90% (a few close to 90%) of simulated neonates and young infants (Fig. 5a). The results indicated that a dose of at least 14 and 15 mg/kg every 12 h was required to attain the target AUC0–24 of ≥ 400 in 90% of infants with a PMA of 30–32 and 32–34 weeks, respectively. This target was also achieved in 93% of simulated infants in the oldest PMA groups (36–38 and 38–40 weeks, respectively) when the dosing interval was extended to 8 h. For infants with a PMA ≥44 weeks, an increase in the dose to 18 mg/kg every 8 h was needed. The trough concentrations corresponding to an AUC0–24 of ≥400 were 5–15 mg/L (Fig. 5b).
Fig. 5

Monte Carlo simulation analysis examining AUC0–24 and trough achievement in neonates and young infants. a AUC-guided dosing regimen achieved AUC0–24 of ≥400 in ≥90% (a few close to 90%) of patients. b Trough concentration corresponding to an AUC0–24 of ≥400

Discussion

In our study, vancomycin was well described by a one-compartment model. This is explained in part by the fact that the available data were scarce, since they derived from routine TDM. For a patient population of neonates and infants, growth (body mass) and maturation (age) are linked highly-related processes which may both influence the PK parameters [27]. In this case, it is reasonable to first use allometric scaling to account for the influence of body size, and to then conduct a covariate analysis using age-related factors to explore the impact of maturation on PK parameters [28]. Although serum creatinine levels in the first few days of life reflect maternal levels more than neonatal renal function [29], based on principles of developmental pharmacology, covariates should include factors of size, maturational processes that can affect drug transporters, and factors affecting kidney function [18]. In our study, BW, PMA and Scr were highly significant in the model. The typical population parameter estimate of clearance for an infant weighing 2.73 kg with a PMA of 39.8 weeks and serum creatinine of 0.28 mg/dL was 0.103 L/h/kg. This is very similar to the values of 0.110 and 0.118 L/h/kg observed by Frymoyer et al. and Capparelli et al., respectively [14, 30]. The distribution volume found in our study (0.58 L/kg) was also similar to that in previous reports [23, 30]. Both internal and external evaluation supported the prediction of the final vancomycin PK model.

However, in the external evaluation, the population prediction errors were a bit large. This can be explained by various factors. First, the between-subject variability in clearance was still significant (ωCL = 0.27) after inclusion of weight, PMA and serum creatinine. Studies have postulated an impact of protein binding and transporters on the large variability in the extent of renal clearance [31, 32]. In addition, the unexplained variability remained significant, as the residual variability was 23.9%. This is likely because the data were based on routine TDM.

A recent multicenter retrospective study by Ringenberg et al. [33] investigated the achievement of therapeutic vancomycin trough serum concentrations (Ctrough) with dosing guidelines from the NeoFax® guide. Of the 171 vancomycin serum trough concentrations, only 25% of the neonates studied achieved a target Ctrough of 10–20 mg/L with empiric dosing. Similarly, in our hospital, based on Neofax® dosing recommendations, we found that only 12.7% of neonates and young infants had troughs in the range of 10–20 mg/L, and 30% were less than 5 mg/L, with most in the range of 5–10 mg/L, which is consistent with our local trough target. Although an increased Ctrough is usually associated with a higher AUC0–24, it is not always a good predictor of the AUC0–24, as it is highly dependent on the dosing interval. Daily doses with longer dosing intervals were found to result in lower troughs. Therefore, dose adjustments could not be predicted with precision for a young infant based solely on Ctrough values. This highlights the importance of considering the AUC0–24 and MIC when treating invasive MRSA infections.

Effective vancomycin dosing is essential to providing optimal treatment and limiting the spread of resistance. Higher vancomycin doses have been recommended for invasive MRSA infections in adults to maximize the odds of achieving an AUC0–24/MIC of ≥400 [9]. Most hospitals have developed local dosing recommendation in neonates. In the present study, the vancomycin model that was developed was then used to simulate dosing regimens in order to determine which of these regimens could reliably generate the targeted AUC0–24. This allows for more precise estimation of vancomycin exposure against the MIC, providing for more accurate dose adjustments to optimize vancomycin treatment. For neonates and young infants in our hospital, attempts to increase vancomycin doses may not result in additional clinical benefit and may increase the likelihood of toxicity. But it is clear that to achieve the adequate AUC0–24/MIC target, the dosage of vancomycin should be increased if bacterial MIC increases. The trough concentrations corresponding to an AUC0–24 of ≥400 were also examined across neonates and young infants of different sizes, developmental ages and serum creatinine levels. Our study suggests that Ctrough values of 5–15 mg/L are highly predictive of an AUC0–24 ≥400 when treating invasive MRSA infections with an MIC of ≤1 mg/L, while the Ctrough of 15–20 mg/L, which is recommended in adults, is not necessary to achieve this target. This understanding would be helpful for framing target trough concentrations in this population, in whom MRSA infection is a concern.

One limitation of our study is that all the data were derived from a single center, which could limit the generalizability of the results. However, the clinical benefits of personalized medical treatment for newborns are clear: the exposure target can be reached earlier, and the number of TDM samples can be potentially reduced. The next step is to encourage other institutions to validate our final model at their institution with their patient population, enabling a better understanding of vancomycin exposure in each individual patient to allow for dose optimization. In addition, the population model relied on 1–2 samples to determine the therapeutic AUC0–24 target. It is critically important to collect a sufficient number of samples to reliably calculate vancomycin AUC0–24.

Conclusion

In summary, vancomycin PK was well described by a one-compartment model, with size, PMA and renal function as significant covariates, in neonates and young infants. The final model was also evaluated in an additional 57 neonates and infants, and showed accurate predictive performance. The information on serum trough concentration alone is not enough to predict vancomycin exposure for this population. The model developed in this work can be used in hospitals to predict individualized vancomycin dosing regimens for patient populations with similar characteristics based on AUC0–24/MIC.

Supplementary material

228_2018_2454_MOESM1_ESM.docx (636 kb)
ESM 1 (DOCX 635 kb)

References

  1. 1.
    Clark RH, Bloom BT, Spitzer AR, Gerstmann DR (2006) Reported medication use in the neonatal intensive care unit: data from a large national data set. Pediatrics 117(6):1979–1987CrossRefPubMedGoogle Scholar
  2. 2.
    Jacqz-Aigrain E, Zhao W, Sharland M, van den Anker JN (2013) Use of antibacterial agents in the neonate: 50 years of experience with vancomycin administration. Semin Fetal Neonatal Med 18(1):28–34CrossRefPubMedGoogle Scholar
  3. 3.
    Kearns GL, Abdel-Rahman SM, Alander SW, Blowey DL, Leeder JS, Kauffman RE (2003) Developmental pharmacology-drug disposition, action, and therapy in infants and children. N Engl J Med 349(12):1157–1167CrossRefPubMedGoogle Scholar
  4. 4.
    Stockmann C, Roberts JK, Yu T, Constance JE, Knibbe CA, Spigarelli MG, Sherwin CM (2014) Vancomycin pharmacokinetic models: informing the clinical management of drug-resistant bacterial infections. Expert Rev Anti-Infect Ther 12(11):1371–1388CrossRefPubMedGoogle Scholar
  5. 5.
    Rybak M, Lomaestro B, Rotschafer JC, Moellering R Jr, Craig W, Billeter M, Dalovisio JR, Levine DP (2009) Therapeutic monitoring of vancomycin in adult patients: a consensus review of the American Society of Health-System Pharmacists, the Infectious Diseases Society of America, and the Society of Infectious Diseases Pharmacists. Am J Health Syst Pharm 66(1):82–98CrossRefPubMedGoogle Scholar
  6. 6.
    Rybak MJ (2006) Pharmacodynamics: relation to antimicrobial resistance. Am J Med 119(6 Suppl 1):S37–S44CrossRefPubMedGoogle Scholar
  7. 7.
    Rybak MJ (2006) The pharmacokinetic and pharmacodynamic properties of vancomycin. Clin Infect Dis 42(Suppl 1):S35–S39CrossRefPubMedGoogle Scholar
  8. 8.
    Craig WA (2003) Basic pharmacodynamics of antibacterials with clinical applications to the use of beta-lactams, glycopeptides, and linezolid. Infect Dis Clin N Am 17(3):479–501CrossRefGoogle Scholar
  9. 9.
    Liu C, Bayer A, Cosgrove SE, Daum RS, Fridkin SK, Gorwitz RJ, Kaplan SL, Karchmer AW, Levine DP, Murray BE, J Rybak M, Talan DA, Chambers HF, Infectious Diseases Society of America (2011) Clinical practice guidelines by the infectious diseases society of America for the treatment of methicillin-resistant Staphylococcus aureus infections in adults and children. Clin Infect Dis 52(3):e18–e55CrossRefPubMedGoogle Scholar
  10. 10.
    Camaione L, Elliott K, Mitchell-Van Steele A, Lomaestro B, Pai MP (2013) Vancomycin dosing in children and young adults: back to the drawing board. Pharmacotherapy 33(12):1278–1287CrossRefPubMedGoogle Scholar
  11. 11.
    Gordon CL, Thompson C, Carapetis JR, Turnidge J, Kilburn C, Currie BJ (2012) Trough concentrations of vancomycin: adult therapeutic targets are not appropriate for children. Pediatr Infect Dis J 31(12):1269–1271CrossRefPubMedGoogle Scholar
  12. 12.
    Frymoyer A, Guglielmo BJ, Hersh AL (2013) Desired vancomycin trough serum concentration for treating invasive methicillin-resistant staphylococcal infections. Pediatr Infect Dis J 32(10):1077–1079CrossRefPubMedGoogle Scholar
  13. 13.
    Lanke S, Yu T, Rower JE, Balch AH, Korgenski EK, Sherwin CM (2017) AUC-guided vancomycin dosing in adolescent patients with suspected sepsis. J Clin Pharmacol 57(1):77–84CrossRefPubMedGoogle Scholar
  14. 14.
    Frymoyer A, Hersh AL, El-Komy MH, Gaskari S, Su F, Drover DR, Van Meurs K (2014) Association between vancomycin trough concentration and area under the concentration-time curve in neonates. Antimicrob Agents Chemother 58(11):6454–6461CrossRefPubMedPubMedCentralGoogle Scholar
  15. 15.
    Le J, Bradley JS, Murray W, Romanowski GL, Tran TT, Nguyen N, Cho S, Natale S, Bui I, Tran TM, Capparelli EV (2013) Improved vancomycin dosing in children using area under the curve exposure. Pediatr Infect Dis J 32(4):e155–e163CrossRefPubMedPubMedCentralGoogle Scholar
  16. 16.
    Hahn A, Frenck RW Jr, Zou Y, Vinks AA (2015) Validation of a pediatric population pharmacokinetic model for vancomycin. Ther Drug Monit 37(3):413–416CrossRefPubMedPubMedCentralGoogle Scholar
  17. 17.
    Janssen EJ, Välitalo PA, Allegaert K, de Cock RF, Simons SH, Sherwin CM, Mouton JW, van den Anker JN, Knibbe CA (2015) Towards rational dosing algorithms for vancomycin in neonates and infants based on population pharmacokinetic modeling. Antimicrob Agents Chemother 60(2):1013–1021CrossRefPubMedGoogle Scholar
  18. 18.
    Samardzic J, Allegaert K, Wilbaux M, Pfister M, van den Anker JN (2016) Quantitative clinical pharmacology practice for optimal use of antibiotics during the neonatal period. Expert Opin Drug Metab Toxicol 12(4):367–375CrossRefPubMedGoogle Scholar
  19. 19.
    Sheiner LB, Rosenberg B, Melmon KL (1972) Modelling of individual pharmacokinetics for computer-aided drug dosage. Comput Biomed Res 5(5):411–459CrossRefPubMedGoogle Scholar
  20. 20.
    Schwartz GJ, Feld LG, Langford DJ (1984) A simple estimate of glomerular filtration rate in full-term infants during the first year of life. J Pediatr 104(6):849–854CrossRefPubMedGoogle Scholar
  21. 21.
    Thomson Reuters clinical editorial staff (2011) Neofax 2011, 24th ed.Google Scholar
  22. 22.
    Marqués-Miñana MR, Saadeddin A, Peris JE (2010) Population pharmacokinetic analysis of vancomycin in neonates. A new proposal of initial dosage guideline. Br J Clin Pharmacol 70(5):713–720CrossRefPubMedPubMedCentralGoogle Scholar
  23. 23.
    Anderson BJ, Allegaert K, Van den Anker JN, Cossey V, Holford NH (2007) Vancomycin pharmacokinetics in preterm neonates and the prediction of adult clearance. Br J Clin Pharmacol 63(1):75–84CrossRefPubMedGoogle Scholar
  24. 24.
    Ette EI, Williams PJ, Kim YH, Lane JR, Liu MJ, Capparelli EV (2003) Model appropriateness and population pharmacokinetic modeling. J Clin Pharmacol 43(6):610–623CrossRefPubMedGoogle Scholar
  25. 25.
    Comets E, Brendel K, Mentré F (2008) Computing normalised prediction distribution errors to evaluate nonlinear mixed-effect models: the npde add-on package for R. Comput Methods Prog Biomed 90(2):154–166CrossRefGoogle Scholar
  26. 26.
    van der Meer AF, Marcus MA, Touw DJ, Proost JH, Neef C (2011) Optimal sampling strategy development methodology using maximum a posteriori Bayesian estimation. Ther Drug Monit 33(2):133–146PubMedGoogle Scholar
  27. 27.
    Vinks AA (2011) Important role of population pharmacokinetic/pharmacodynamic modeling in pediatric therapeutics. J Pediatr 159(3):361–363CrossRefPubMedGoogle Scholar
  28. 28.
    Anderson BJ, Holford NH (2011) Tips and traps analyzing pediatric PK data. Paediatr Anaesth 21(3):222–237CrossRefPubMedGoogle Scholar
  29. 29.
    Vieux R, Hascoet JM, Merdariu D, Fresson J, Guillemin F (2010) Glomerular filtration rate reference values in very preterm infants. Pediatrics 125(5):e1186–e1192CrossRefPubMedGoogle Scholar
  30. 30.
    Capparelli EV, Lane JR, Romanowski GL, McFeely EJ, Murray W, Sousa P, Kildoo C, Connor JD (2001) The influences of renal function and maturation on vancomycin elimination in newborns and infants. J Clin Pharmacol 41(9):927–934CrossRefPubMedGoogle Scholar
  31. 31.
    Golper TA, Noonan HM, Elzinga L, Gilbert D, Brummett R, Anderson JL, Bennett WM (1988) Vancomycin pharmacokinetics, renal handling, and nonrenal clearances in normal human subjects. Clin Pharmacol Ther 43(5):565–570CrossRefPubMedGoogle Scholar
  32. 32.
    Marsot A, Boulamery A, Bruguerolle B, Simon N (2012) Vancomycin: a review of population pharmacokinetic analyses. Clin Pharmacokinet 51(1):1–13CrossRefPubMedGoogle Scholar
  33. 33.
    Ringenberg T, Robinson C, Meyers R, Degnan L, Shah P, Siu A, Sturgill M (2015) Achievement of therapeutic vancomycin trough serum concentrations with empiric dosing in neonatal intensive care unit patients. Pediatr Infect Dis J 34(7):742–747CrossRefPubMedGoogle Scholar

Copyright information

© Springer-Verlag GmbH Germany, part of Springer Nature 2018

Authors and Affiliations

  • Yewei Chen
    • 1
  • Dan Wu
    • 1
  • Min Dong
    • 2
  • Yiqing Zhu
    • 1
  • Jinmiao Lu
    • 1
  • Xiaoxia Li
    • 1
  • Chao Chen
    • 3
  • Zhiping Li
    • 1
  1. 1.Department of PharmacyChildren’s Hospital of Fudan UniversityShanghaiChina
  2. 2.Division of Clinical PharmacologyCincinnati Children’s Hospital Medical CenterCincinnatiUSA
  3. 3.Department of NeonatologyChildren’s Hospital of Fudan UniversityShanghaiChina

Personalised recommendations